The American Bid Vanished: US Traders Sold Bitcoin at a Loss This Week While the Rest of the World Bought
Bitcoin spent Thursday doing the one thing nobody expected it to do this week: nothing dramatic. After a failed Senate vote on Tuesday and the first Federal Reserve rate hike since 2023 on Wednesday, the largest cryptocurrency traded around $76,300, roughly 1% higher over 24 hours, and comfortably inside the $75,000–$77,000 band it has occupied since Monday.
The price held. The buyers did not.
America stopped paying up
The cleanest measure of American demand is the gap between bitcoin’s dollar price on Coinbase and its USDT price on Binance. CryptoQuant’s Coinbase Premium Index — that spread expressed as a percentage — printed around -0.07% to -0.079% on Tuesday, its lowest reading since mid-August. In cash terms it is a thin margin, roughly $50 of discount on a coin trading near $75,900. But direction matters more than size: the premium had turned positive for the first time in months in late August and early September as bitcoin pushed toward $80,000. It has now flipped back, and it flipped on the day the CLARITY Act died.
On-chain analyst Willy Woo framed the same divergence in cumulative volume delta, which tracks whether aggressive buying or selling dominates on each venue. Since around September 11, Binance’s CVD has been climbing while Coinbase’s kept falling. His read: US-based sellers led, offshore accumulation continued. That interpretation is worth holding loosely — CVD measures aggressive flow by venue, not the nationality of the humans behind it, and a market maker hedging on one exchange can look like a seller there and a buyer somewhere else.
23,200 coins moved at a loss
The wallets that did move were the newest ones. CryptoQuant data shows short-term holders — coins held under six months — pushing exchange inflows from about 19,400 BTC to 33,100 BTC in a rolling 24-hour window, a 71% jump. Of that, 23,200 BTC went to exchanges while underwater, about $1.79 billion of coins moved at a loss. It was the largest short-term-holder capitulation event in a month.
The distribution was telling. Kraken took in more than 6,000 BTC, well above its usual 2,000–3,000 range. Binance absorbed more than 10,000. Coinbase Advanced received 7,300, close to its normal range — so this was not a single venue or a single whale, it was a cohort of recent buyers deciding that $75,000 was not the price they had signed up for.
Their cushion is thinning fast. Short-term holders’ average cost basis sits near $71,300, which means the group’s aggregate profit margin has compressed from about 10% to 7% in a week. CryptoQuant’s exchange-inflow profit/loss indicator has collapsed roughly 96% since September 10 and is hovering near zero, the point where coins arriving at exchanges at a profit are matched by coins arriving at a loss.
Four gauges went quiet at the same time
Glassnode’s Week Onchain newsletter, published this week, is blunter than the price action. It describes a market where the capital that funded the August rally simply stopped arriving:
- Realized Cap, the aggregate value of every coin at the price it last moved, rose for 27 consecutive days through September 14 — then printed its first negative daily reading on September 15. It currently sits near $1.069 trillion.
- US spot bitcoin ETFs have bled roughly $1.05 billion across six of the seven sessions from September 8 to 16. Tuesday’s -$450.3 million was the heaviest single-day redemption since June 25; Wednesday added -$295.9 million, bringing the two-day total to about $746 million. BlackRock’s IBIT alone gave back $144.1 million on Wednesday, ARKB $84.4 million, FBTC $52.7 million. Ethereum funds lost a combined $365.6 million over the same two sessions.
- Stablecoin supply is flat on the week at roughly $301 billion, about 4% below its April peak — the dry powder for the next leg is not being minted.
- Corporate treasuries bought roughly 5,900 BTC in three months, against 89,000 BTC in July 2025 alone.
That last one carries a number worth memorising: the average corporate treasury entry sits at $80,500, about 6% above spot. The cohort that spent 2025 being treated as structural support is now under water and no longer adding. Glassnode’s verdict is unsentimental — “a buyer that has stopped buying and holds a paper loss is not support” — and price has twice tried and failed to reclaim that line in 2026, in May and again on September 3.
What the levels say
Bitcoin is trading about 1% below its True Market Mean of $76,700, the average price paid by active investors. Glassnode’s framework is binary: two consecutive daily closes back above that line, paired with Realized Cap turning positive again, restores the old range and puts $80,500 back in play. A second close below it confirms the break, and the near bids are the short-term-holder cost basis at $71,300, then the $62,000–$65,000 zone where deeper accumulation happened. Order-book liquidity thins noticeably below $68,000.
Prediction-market contracts, which are indicative rather than predictive, currently price roughly a 65% chance bitcoin finishes Friday above $76,000 and about 97% that it holds $74,000 on Saturday. In other words, the crowd expects a range, not a resolution.
Two caveats on the bearish reading. First, ETF flow data does not identify who redeemed or why — a redemption can be hedging, rotation or a treasury desk squaring books, and one week does not establish a trend. Second, derivatives positioning improved even as spot demand faded: Thursday’s $345 million liquidation wave, across 86,816 traders, was $208 million shorts against $137 million longs. Traders positioned for a breakdown got squeezed instead. That is forced buying rather than conviction, but it is not distribution either.
The uncomfortable part of this week is not that bitcoin fell. It is that bitcoin barely fell while nearly every channel that delivered demand in 2025 went quiet at once — and nobody yet knows whether that is a pause or a handover.
This article is for information only and does not constitute investment advice. Figures reflect the most recent data available at publication.